Gamal Seedahmed 
  
marks are registered onto the film by projecting an image of the mark through a small lens housed 
inside the cone (Schenk, 1999). Since each fiducial is designed and projected separately, small 
variations may occur. 
Fiducial marks have typical patterns, made up of geometrical structures including straight-line 
segments, crosses, solid squares and circles, and annulus, see Fig. (1). Most of the structures serve the 
identification process and lend themselves as suitable candidates for a Hough Transform. The actual 
fiducial center, whose coordinate is known from camera calibration, is a small disc, just slightly bigger 
than the measuring mark of the analytical plotter (Schenk, 1999). 
  
NR 
YS WEA 
yw x / 
" e NS 
Fiducial Image CAD Drawing of Fiducial 
Fig. 1: A portion of a fiducial image and its CAD design provided by the manufacturer. 
1.3 Overview of Automatic Interior Orientation 
The automation of the interior orientation needs to concentrate primarily on recognizing and measuring 
the fiducial marks. In general there are many different ways to automatically or semi-automatically 
locate the fiducials on the digital images, such as manual identification or providing an approximate 
position of the fiducial and then determine its center by using an automatic mensuration technique e. g. 
mathematical morphology or fully automatic fiducial identification and mensuration (Lue, 1997). 
Interior orientation in most existing softcopy workstations requires at least one or two fiducials to be 
measured manually before the remaining fiducials can be automatically determined. In general, interior 
orientation is approached as a template-matching problem augmented with least squares matching 
(Lue, 1997) or with a modified Hough Transform for rough localization (Kersten and Haering, 1997). 
Also interior orientation is solved via binary correlation with a hierarchical approach for rough 
localization and gray level correlation for subpixel accuracy (Schickler and Poth, 1996). 
In this study, we benefit from the fact that fiducial marks are artificial objects projected onto the film 
during the exposure time. Based on this fact, the interior orientation problem is approached as a model- 
based object recognition task utilizing most of the available prior knowledge regarding the fiducials, in 
terms of providing the CAD design data, their shape, their location, the camera optics e. g. the 
projection factor of the fiducials and the pixel size of the image. At the identification level the interior 
orientation is solved as an object-recognition problem, and at the precise localization level it is solved 
as an ordinary least-squares adjustment problem. 
1.4 Objective of Autonomous Interior Orientation 
A system is autonomous to the extent that its behavior is determined by its own experience. In the 
context of this research and our view of digital photogrammetry we are going to use Schenk's (1999) 
definition of autonomous “as if the process is really 100% automatic and does not require human 
intervention- sort of a black-box- then the word autonomous can be used to describe the system”. 
The objective of autonomous interior orientation can be summarized in the following points. The main 
objective is the identification and approximate localization, subpixel localization of fiducial centers, 
and computation of the transformation parameters between the pixel coordinate system and the photo 
coordinate system. Identification includes the task of determining which of the fiducial marks is 
  
824 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.